Enhancing Cancer Diagnostics with AI

Client Overview

The customer is a leading provider of cancer-focused testing and information services, offering a comprehensive range of diagnostic, prognostic, and predictive testing services, including both genetic and molecular diagnostic tests.

Challenge

The customer sought to build a new AI system that would significantly enhance the efficiency and accuracy of pathology workflows. The company needed a solution that could streamline the analysis of advanced diagnostic test results while accurately standardizing diagnoses across observers.

The primary goals were to:

• Allow users to focus on complex cases requiring expert judgment
• Improve overall productivity in cancer diagnostics
• Ensure consistent interpretation of genetic test results
• Reduce time spent per case while maintaining diagnostic accuracy

Solution

Caylent designed and implemented an AI-powered solution on AWS focused on specific types of disease that were key focus areas for the customer. The solution leveraged several key AWS technologies:

AI-Powered Test Panel Analysis System

• Automated Initial Test Panel Analysis: A system that rapidly processes and interprets genetic test data, highlighting key findings for pathologists
• Predictive Modeling: Machine learning capabilities that identify patterns and predict potential outcomes or disease progression based on previous reports and white papers
• Knowledge Graph Implementation: Structured representation of medical knowledge to enhance context-aware analysis

Retrieval Augmented Generation (RAG) Architecture

• Amazon S3 Storage: Secure storage for sensitive diagnostic data
• Large Language Model Integration: Selection of appropriate LLMs based on security, cost, and performance requirements for medical text summarization
• Data Cleansing & Segmentation: Processing pipeline to prepare clinical data for AI analysis

Comprehensive Evaluation Framework

• SageMaker Notebook Environment: For data scientists to develop and refine models
• Streamlit Application: User interface allowing pathologists to interact with and validate the system's outputs
• Testing Framework: Comparing AI-generated results with ground truth data from existing labeled pathology reports
• "Lab in the Loop" Design: Framework for continuous improvement based on pathologist feedback

Results

The implementation delivered significant benefits to diagnostic operations:

1. Enhanced Diagnostic Efficiency: Pathologists reported spending less time per case while maintaining diagnostic accuracy, as the AI system handled initial analysis of test results
2. Improved Standardization: The solution helped standardize interpretations across different observers, ensuring consistent reporting of cancer diagnostic findings
3. Focus on Complex Cases: By automating mundane tasks, pathologists could dedicate more time to complex cases that required their specialized expertise
4. Successful Validation: The evaluation framework demonstrated high confidence in the system's outputs, with pathologists confirming that minimal rewrites were needed for AI-generated summaries
5. Foundation for Future Development: The project established a solid foundation for future phases, including the "Lab in the Loop" feedback system for continuous improvement

The solution specifically addressed the unique requirements of cancer diagnostics, focusing on hematologic cancer and myeloid disorders panels. By implementing this AI system, the customer strengthened their position as a leader in diagnostic testing and information services, enhancing their ability to deliver accurate and timely diagnostic results to patients and healthcare providers.

The project demonstrated how AWS technologies can be effectively applied to life sciences challenges, creating tangible improvements in specialized medical fields like diagnostics and pathology.